This paper introduces a matrix formulation in the compressive spectral imaging sensing problem which reduces computational complexity and allows to design sensing matrices using principal components analysis, leading to improved reconstructions.